JP4126843B2 - Data management method and apparatus, and recording medium storing data management program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a data management technique, and more particularly to a technique that is effective when applied to managing a plurality of data with respect to a key of an index.
[0002]
[Prior art]
With the rapid spread of business systems using the Internet in recent years, the field of application of database management systems (DBMS) that support the system is expanding. In addition, the amount of data handled by DBMS is increasing year by year. There are many applications in the system that use B-tree indexes, which are the most representative high-speed search means of DBMS, for narrow-range data such as “business transition status” in workflows, and the stability of B-tree indexes. It is important to provide the required performance.
[0003]
Regarding the B-tree index, for example, in the literature (Jim Gray and Andreas Reuter, Transaction Processing: Concepts and Techniques, Morgan Kaufmann Publishers, 1993) Is disclosed.
[0004]
In order to realize a high-speed B-tree index as a real system, it is necessary to solve the following problems.
[0005]
(I) Improvement of concurrency by multiple processes
Problems arise when multiple transactions access a database table that contains records simultaneously. Specifically, a contention situation occurs when a transaction attempts to update one record and another transaction attempts to access the same record at the same time. As a solution to the contention problem, there is a locking method (exclusive access method) for a part of a record or a B-tree index (the entire B-tree index, node, index entry, key value, etc.). A locking scheme forces a transaction to acquire a lock before accessing data. At this time, if the lock is applied by another transaction, the lock may collide, so that a necessary lock may not be acquired. The locking method guarantees the update result and reference result of the own transaction, but while acquiring the lock, it waits for the access of other transactions until the lock is released, thus increasing the system multiplicity (simultaneous execution) and throughput. In order to increase the number of locks, it is important to minimize the number of locks acquired simultaneously by one transaction and the influence range (granularity) of the locks.
[0006]
(Ii) Recovery processing control
In an actual system, it is necessary to guarantee the atomicity of a transaction and guarantee the consistency of data in a B-tree index with respect to various failures such as transaction interruption, system down, and media failure. Therefore, the log record acquisition method and the recovery process control using the log record are important. As a log record acquisition method, there is a physical log method in which a physical image before and after updating a node constituting an index is acquired by using log record information. In recovery using a physical image, other transactions are restricted from accessing the page until the updated transaction is completed. In other words, recovery process control is closely related to exclusive control and also affects concurrency.
[0007]
FIG. 11 shows the structure of a conventional typical B-tree index.
[0008]
In the B-tree index, nodes are branched over a number of levels from the root node, with one root node as a vertex. The innermost node of the branched node is often called a leaf node. Nodes other than leaf nodes including the root node are often referred to as upper nodes. Each node has a plurality of index entries composed of the following information. The index entry in the leaf node is composed of a pointer indicating a record (table data) in the database and a key value representing one of the characteristics of the record (table data). The index entry in the upper node is a pointer indicating the node (child node) at the next lower level, and one key value indicating the range of key values that are branched from the child node and are finally managed by the leaf node. Consists of The key value in the higher-order index entry is a guideline when the access program to the B-tree index traces from the root node to the leaf node in which the index entry including the pointer to the target record is stored and managed (determination Element). Normally, one node of the B-tree index is realized by a page which is an access unit on the database.
[0009]
[Problems to be solved by the invention]
One of the exclusive control methods for the B-tree index is a key value exclusion method or a key range exclusion method in which the granularity of the exclusive resource of the index is “key”. When exclusive control for these index keys is used, search / update processing for data (rows) having the same key is serialized by index key exclusion, and concurrency decreases. Transactions in the WEB environment that are currently spreading rapidly are longer than the length of conventional OLTP transactions, and it is difficult to provide high-throughput services with the exclusive control method for keys.
[0010]
Also, as shown in FIG. 11, there are the following problems with respect to the B-tree index having a large number of identical keys. In the B-tree index of FIG. 11, table data information related to a certain key is managed by one leaf entry 16. The leaf entry 16 has a key 14, table data information 18 (data record identifier) related to the key value, and the number 17 (super plural) of information 18 related to the table data related to the key value. The table data information 18 is arranged in ascending order in the form of a plurality of duplicates. In such a storage method, when the number of overlapping for a certain key increases, the length of the leaf entry itself cannot be accommodated in the leaf node 12. As a result, the table data information for one key must be stored in a plurality of nodes as a plurality of leaf entries. In the example of FIG. 11, the entry of the key “28” is divided and stored in three nodes, leaf nodes N4, N5, and N6. In this structure, the entry of the key “28” becomes a bottleneck in the scan using the horizontal pointer of the leaf leaf node to cope with the range search. Furthermore, the key 14 included in the upper entry 13 stored in the root node 10 and the upper node 13 must be a key in the table data information in addition to “28”. The number of the root node 10 and the upper node 13 It is a factor that increases. As a result, the duplicate key causes not only the access process for the duplicate key itself but also the access performance and storage efficiency of the entire B-tree index.
[0011]
Gray discloses a method of managing table data information in a data structure outside the B-tree structure, but does not disclose a concurrent execution control and recovery control method for the method. Further, although a recovery processing method using a physical log is disclosed, simultaneous execution on the same page as described above cannot be realized. Such a problem generally occurs when performing index access to the same key.
[0012]
An object of the present invention is to improve the above-described problem and provide a technique capable of suitably performing index access to the same key in an environment where a plurality of accesses are performed.
[0013]
[Means for Solving the Problems]
The above problems are improved by the following means.
[0014]
In a method of managing a large number of B-tree indexes having the same key in an environment accessed by a plurality of users at the same time, the data structure stores a plurality of data pointers related to a single key, and an index entry including the key is The data structure is different from the stored leaf node. The data structure is pointed to by the index entry containing the key, and has a data structure that can be accessed simultaneously by multiple processes. When adding or deleting, a first log record including information for adding or deleting the data pointer is acquired for the data structure storing a plurality of data pointers related to the key, and is stored in the leaf node. Update the stored index entry containing the key To obtain a second log record containing information indicating that the addition or deletion of data pointer to a structure is made, in the data structure, to improve the above problems by adding or deleting the data pointer.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
A database processing system according to an embodiment capable of efficiently performing index access to the same key in a simultaneous access environment from a plurality of users will be described.
[0016]
First, the concept of the present invention will be briefly described with reference to FIG.
[0017]
A B-tree index 1 in the database management system according to the present embodiment includes a B-tree structure 101 for efficiently acquiring information related to table data using a key as shown in FIG. It consists of a data structure 102 (hereinafter referred to as “duplicate key data structure”) for managing a large number of associated table data information.
[0018]
The B-tree structure 101 includes a root node 10, an intermediate node 11, and a leaf node 12, and has a structure in which nodes are branched from a root node over a number of levels with one root node as a vertex. The root node 10 and the intermediate node 11 store an upper entry 13 having a key 14 and a pointer 13 indicating a node to a lower level related to the key value. Within each node, the upper entry 13 and the leaf entry 16 are stored and managed in the order of the key values of the index entries 13 and 16. In this embodiment, key values are stored in ascending order from left to right in the figure. By performing a binary search on the sorted index entries in each node, the necessary index entries can be accessed efficiently and stably. The child node pointed to by one higher index entry stores an index entry having a key value smaller than the key value in the higher index entry.
[0019]
When there are a large number of table data related to the key 14, the leaf entry 20 is stored in the key 14, the number 17 of information related to the table data related to the key value, and the duplicate key data structure 102 storing information related to the table data. It has a pointer 19. In the example of FIG. 1, the number of related table data information, that is, the overlap number is “35,678” for the key value “28”. The 35,678 pieces of information are stored and managed in the duplicate data structure 102 pointed to by the pointer 19 “Nx1”.
[0020]
On the other hand, in the example of FIG. 1, like the index entry other than the key value “28”, when the number of information regarding the duplication number, that is, the table data is not so large, it is stored as the leaf entry 16 having no duplicate key data structure 102. The leaf entry 16 includes a key 14, information 18 (data record identifier) related to table data related to the key value, and the number 17 of information 18 related to table data related to the key value. When the overlap of the table data information related to the key exceeds a threshold value due to the addition or update of the data record, the processing shifts to the leaf entry 20 that points to the duplicate key data structure 102.
.
[0021]
The duplicate key data structure 102 for storing and managing information related to a large number of table data has the following characteristics.
[0022]
(1) Simultaneous access by multiple processes is possible without acquiring exclusive access.
[0023]
(2) Cancel processing is possible without restricting access to normal processing.
[0024]
Here, an arrow 103 indicates a state of additional update processing of the key value “28” and the table data related information “P18” for the B-tree index 1 accompanying the update of the data record. First, in this additional update process, a search is started from the root node 10 (N1) of the B-tree structure 101 using the key value “28”, and the update target key value “28” is passed through the intermediate node 11 (N2). To the leaf node 12 (N5) in which the leaf entry 20 having "" is stored. Then, the leaf entry 20 is confirmed, the first log record 41 is acquired, and the duplication number 17 in the entry is incremented. The first log record 41 includes information necessary for continuing the update process for the duplicate key data structure 102. Further, the pointer 19 (Nx1) to the duplicate key data structure 102 in the entry is acquired, and the duplicate key data structure 102 is accessed. Then, the second log record 42 is acquired, and table data related information “P18” is added to the duplicate key data structure 102. The second log record 42 includes information indicating that the addition of the table data related information 18 to the duplicate key data structure 102 has been completed.
[0025]
Although the addition of table data related information has been described using the example of FIG. 1, the same applies to the deletion of table data related information. The leaf entry 20 is accessed, the first log record 41 is acquired, and the duplication number 17 in the entry is decremented. Then, after accessing the duplicate key data structure 102 and acquiring the second log record, the table data related information 18 is deleted from the duplicate key data structure 102.
[0026]
As described above, the order of change from the update of the leaf entry to the duplicate key data structure is constant, and exclusive control of the data resources constituting the index 1 is not required, so that a plurality of processes can be executed simultaneously. . Further, even when a canceling factor occurs for a series of change processing results or when a processing interrupting factor occurs immediately after updating a leaf entry, the first log record 41 and the second log record 42 described above are used. By doing so, it is possible to carry out cancellation processing without losing concurrency.
[0027]
Specifically, during the cancellation process, if the first log record 41 and the second log record are acquired in pairs, it is determined that a series of change processes have been completed. In the example of FIG. The deletion process of the value “28” and the table data related information “P18” is performed. When only the first log record 41 is acquired at the time of the cancellation process, only the change process for the leaf entry 20 is performed, and therefore the table for the duplicate key data structure 201 using the information included in the first log record. After the addition process of the data related information “P18”, the key value “28” and the table data related information “P18” are deleted again. Even in the cancellation process, the order of changes to the B-tree index 1 is always one-way, so that the concurrency with other processes is not impaired. As described above, high concurrency can be provided even with cancellation processing.
[0028]
Further, since the duplicate key data structure 201 is pointed from the leaf entry 20 and stored and managed outside the leaf node 12, all the table data related information 18 is stored in the leaf node 12 in which the leaf entry 20 is stored. Compared with the system having the leaf entry 16, the leaf entry having more other keys can be stored. Then, the necessary number of leaf nodes 12 for storing leaf entries is reduced, and finally the necessary number of upper nodes 11 is also reduced, thereby reducing the storage capacity of the entire B-tree index. Even for access to a key that does not have much table data information, the capacity of the B-tree structure is suppressed, so that it is possible to provide stable performance for all search processes and change processes that follow the B-tree structure.
[0029]
Next, FIG. 2 shows a schematic configuration of the database management system of the present embodiment. There is an application program 6 created by a user and a database management system 2 that manages the entire database system such as inquiries and resource management. The database management system 2 includes a logical processing unit 21, a physical processing unit 22, a system control 23, a database 3 that permanently or temporarily stores data to be accessed by the database, and a system log 4. The database management system 2 is connected to other systems via a network or the like.
[0030]
The logic processing unit 21 includes a query analysis 211 that performs syntax analysis / semantic analysis of a query, an optimization process 212 that generates an appropriate processing procedure, and a code generation 213 that generates a code corresponding to the processing procedure. ing. Further, a code interpretation execution 214 for interpreting the code and instructing the physical processing unit 22 to execute the database processing based on the generated code is provided.
[0031]
The physical processing unit 22 includes a table data management unit 221 that performs data storage, deletion, update, and search processing on the table data 5 stored in the database 3 via the database buffer 24. In addition, a B-tree index management unit 25 that manages the index 1 stored in the database 3 is provided. The index 1 includes a B-tree structure for efficiently accessing related information using a key, and a duplicate key data structure for managing a large number of table data information associated with the key for each key.
[0032]
As the table data 5 is updated, the B-tree index management unit 25 performs a function of the B-tree structure access processing 26 for changing the B-tree structure for the B-tree index 1 related thereto, and changes to the duplicate data structure. It has the function of the duplicate key data structure access processing 27 to be performed. Further, the B-tree structure access process 26 and the duplicate key data structure access process 27 have a function for searching the related table data at high speed from the search condition key by accessing the index 1.
[0033]
The physical processing unit 22 also includes an exclusive control 223 for resources shared by the system. In this embodiment, data consistency is realized only by exclusive control (exclusive resources: table data storage page ID, table data ID, etc.) for table data 5, and B-tree index 1 resources (index ID, index page ID). , Key value, etc.) are not controlled exclusively.
[0034]
The system log 4 accumulates update history information such as data insertion, update, and deletion in the database 3. The accumulated information includes history information regarding changes to the B-tree index 1. It is used to restore data consistency of the B-tree index 1 in various situations such as transaction cancellation and system down.
[0035]
The system control 23 performs inquiry input from the user, return of inquiry results, database maintenance by command reception, and the like. In addition, the system log 4 is managed, and in cooperation with the physical processing unit 22, log records that are history information are acquired 26 when the database 3 is changed. Then, when the read log record relates to a change process for the B-tree index, the system control 23 passes the log record to the B-tree index management unit 25 of the physical processing unit 22 and requests cancellation processing. The requested B-tree index management unit 25 performs change cancellation processing based on the information in the log record.
[0036]
A user can search, access, and change data records constituting the table data 5 in the database 3 using the database management system 2 via the application program 6. The database management system 2 uses the B-tree index 1 to realize efficient access to data records.
[0037]
The B-tree index 1 or the table data 5 in the database 3 can be accessed by reading (GET) it into the database buffer 24 in units of pages which are access units. After referencing (searching) the page on the buffer, the buffer is released by releasing the buffer, and the use of the buffer from other processing is permitted. Also, after updating the page on the buffer, declare that the page on the buffer will be reflected in the database by PUT the buffer, and allow the buffer to be used from other processing. . The page on the buffer is not immediately reflected in the database, but is reflected in the database at a certain moment. Latch the buffer so that other transactions do not access the page GET to the buffer. Although a latch is a kind of lock, the acquisition period is much shorter than that of a normal lock, and acquisition and release can be performed at a low cost. In this embodiment, the shared latch is applied when referring to the index node, and the exclusive latch is applied when updating. Processing related to the exclusive latch is serialized, and simultaneous references to the buffer are possible between the shared latches. Then, the latch being acquired is released when the buffer is released, and access to another process that is trying to acquire the exclusive latch is permitted.
[0038]
FIG. 3 is a diagram illustrating an example of a hardware configuration of the computer system according to the present embodiment. The computer system 3000 includes a CPU 3002, a main storage device 3001, an external storage device 3003 such as a magnetic disk device, and a number of terminals 3004. The database management system 2 described above with reference to FIG. 2 is placed on the main storage device 3001, and the database 3 including the table data 5 and the index 1 managed by the database management system 3 is placed on the external storage device 3003. Stored. The system log 4 is also stored on the external storage device 3003. Further, a program 3100 for realizing the database management system 2 is also stored on the external storage device 3003.
[0039]
FIG. 8 is a diagram showing an example of an index log record according to the present embodiment.
[0040]
As shown in 41 of FIG. 8, the first log record 41 acquired when the leaf entry described in FIG. 1 is changed is a log record header 410 including information of a transaction to which the change process belongs, an operation indicating the change A code 411, a key 412, table data information 413 to be changed, and a pointer 414 to a duplicate key data structure pointed to by the changed leaf entry. In the operation code 411, a code indicating an addition process or a deletion process is set. In the example of FIG. 8, the setting code is “INSERT” indicating addition processing. The log record header 410 includes an index ID of the B-tree index to be changed and area information for storing the index. The value set in the key 412 is the same as the value of the key 14 in the leaf entry 20 to be changed. In the example of FIG. 8, the value is “28”.
[0041]
The pointer 414 to the duplicate key data structure continues the change process for the duplicate key data structure when a change result canceling factor occurs in a situation where the leaf entry change is completed but the duplicate key data structure change is not completed. It is information for.
[0042]
Further, as indicated by 42 in FIG. 8, the second log record acquired when the leaf entry described in FIG. 1 is changed is a log record header 420 including information on the transaction to which the change process belongs, and an operation indicating the change. It consists of a code 421 and table data information 422 to be changed. Similar to 411, the operation code 421 is set with a code indicating addition processing or deletion processing. In the example of FIG. 8, the setting code is “INSERT” indicating addition processing. Further, the same value as 413 of the first log record 41 is set in the table data information 422. As described above with reference to FIG. 1, the second log record 42 serves as a marker that indicates whether the change process for the duplicate key data structure has been completed.
[0043]
FIG. 4 is a flowchart showing a processing procedure of the B-tree index management unit 25 of this embodiment. FIG. 4 shows the processing contents of the B-tree structure access processing 26 when table data information is added to the B-tree index.
[0044]
First, in step 401, using the key, the B tree structure is traced from the root node to the upper node, and after determining the leaf node, the leaf node is read into the database buffer and the buffer is fixed (GET). Next, in step 402, the leaf entry stored in the leaf node is searched, and the leaf entry to be changed is determined. In step 403, the number (overlapping number) of table data information related to the key of the leaf entry is incremented. Here, in step 404, a pointer to the duplicate key data structure included in the leaf entry is acquired. Further, the first log record described with reference to FIG. 8 is created (step 405), and the created first log record is acquired (step 406). Finally, the leaf node is reserved to be written in the database, and the buffer fixation is released (PUT) (step 407). Thus, the leaf entry change process is completed (step 408), and the process proceeds to the change process for the duplicate key data structure.
[0045]
In this flowchart, processing is performed in the order of leaf entry change (404) and log record acquisition (405), but the reflection of the change result for the leaf node in the actual database is in accordance with the WAL (Write-ahead log) protocol. This is done after the log record is written to the system log.
[0046]
FIG. 5 is a flowchart showing a processing procedure of the B-tree index management unit 25 of the present embodiment. FIG. 5 shows the processing contents of the duplicate key data structure access processing 27 when table data information is added to the B-tree index.
[0047]
The change process for the duplicate key data structure is continued from the change process for the leaf entry described in the flowchart of FIG.
[0048]
First, in step 501, the duplicate key data structure is accessed using the pointer previously obtained in step 404 of FIG. 4, and in step 502, a position for adding table data information is searched. Then, the second log record described with reference to FIG. 8 is created (step 503), and the created second log record is acquired (step 504). Here, it is determined whether there is an additional area in the duplicate key data structure (step 505). If there is no additional area, the process proceeds to step 506, where a new area is allocated, and the duplicate key data structure is changed when the new area is added (step 507). Specifically, when the duplicate key data structure is composed of a plurality of pages, new page allocation is performed. Thereafter, the process proceeds to step 508. If the determination result in step 505 is that there is a region, the process proceeds to step 508, table data information is added to the duplicate key data structure, and the change process for the duplicate key data structure is terminated (step 509).
[0049]
FIG. 6 is a flowchart showing a processing procedure of the B-tree index management unit 25 of this embodiment. FIG. 6 shows the processing contents of the B-tree structure access processing 26 when table data information is deleted for the B-tree index.
[0050]
First, in step 601, using the key, the B-tree structure is traced from the root node to the upper node, and after determining the leaf node, the leaf node is read into the database buffer and the buffer is fixed (GET). Next, at step 602, the leaf entry stored in the leaf node is searched, and the leaf entry to be changed is determined. Here, in step 603, a pointer that points to the duplicate key data structure included in the leaf entry is obtained. In step 604, it is determined whether or not the number (overlapping number) of table data information related to the key of the leaf entry is one. If the overlap number is 1, the process proceeds to step 605 to delete the leaf entry, and the process proceeds to step 607. If the determination result in step 604 is that the duplication number is greater than 1, the process proceeds to step 606, and the number of table data information (duplication number) related to the key of the leaf entry is decremented.
[0051]
Further, the first log record described with reference to FIG. 8 is created (step 607), and the created first log record is acquired (step 608). Finally, the leaf node is reserved to be written in the database, and the buffer fixation is released (PUT) (step 609). Thus, the leaf entry change process is completed (step 610), and the process proceeds to the change process for the duplicate key data structure.
[0052]
FIG. 7 is a flowchart showing a processing procedure of the B-tree index management unit 25 of the present embodiment. FIG. 7 shows the processing contents of the duplicate key data structure access processing 27 when table data information is deleted for the B-tree index.
[0053]
The change process for the duplicate key data structure is continued from the change process for the leaf entry described with reference to the flowchart of FIG.
[0054]
First, in step 701, the duplicate key data structure is accessed using the pointer previously obtained in step 603 of FIG. 4, and in step 702, the existence position of the deletion target table data information is searched. Then, the second log record described with reference to FIG. 8 is created (step 703), and the created second log record is acquired (step 704).
[0055]
Next, in step 705, the table data information is deleted from the duplicate key data structure. Here, it is determined whether or not an area constituting a recoverable duplicate key data structure is generated by deleting the table data information (step 706). As a result of the determination, if a recoverable area occurs, the process proceeds to step 707, where the area is recovered, and the duplicate key data structure is changed according to the area recovery (step 708). Specifically, when the duplicate key data structure is composed of a plurality of pages, the pages are collected. The collected pages are reused by assigning new pages in other processes. In step 709 after step 708, the process for changing the duplicate key data structure is terminated. If the determination result in step 706 indicates that there is no recoverable area, the process for changing the duplicate key data structure is terminated in step 709.
[0056]
FIG. 9 is a flowchart showing the processing procedure of the index update result cancellation processing of the index management unit of this embodiment.
[0057]
First, in step 901, a log record related to a B-tree index change is received from system control. In step 902, it is determined whether there is a log record to be processed. If there is no log record to be processed, the index update result cancellation process is terminated in step 913. If there is a log record to be processed, that is, if there is a received log record, the process proceeds to step 903 to determine the type of log record. Here, it is first determined whether the log record is the second log record (step 903). If it is the second log record, the process proceeds to step 904 to hold the second log record. Then, the process returns to step 901 to process the next log record. Here, the next log record should be the first log record. If it is determined in step 903 that the log record is not the first log record, it is further determined in step 905 whether the log record is the first log record. If it is not the first log record, update result cancellation processing corresponding to other log records is performed in step 906, and the processing returns to step 901 to process the next log record. If it is the first log record, the following processing is performed. First, it is determined whether or not the second log record is retained (step 907). If retained, it is determined that a series of processing for changing the leaf entry and the duplicate key data structure has been completed and retained. The second log record is discarded (step 908), and the process proceeds to step 910. If the second log record is not held, it is determined that the change processing of the leaf entry has been completed but the change processing for the duplicate key data structure to be paired has not been completed. The update process is continued. Thereafter, the process proceeds to step 910. In step 910, it is determined whether the log record being processed has been acquired by normal processing. If it is a log record obtained by normal processing, it is necessary to cancel the related change result. In step 911, the cancellation process is performed. Further, in the case of acquisition not during normal processing, that is, acquisition during cancellation processing, it is determined that cancellation processing has already been completed, including step 909, and a series of B-tree index update result cancellation processing ends, and step 901 is completed. Return to and process the next log record.
[0058]
FIG. 10 is a diagram showing an example of the configuration of the duplicate key data structure 102 of this embodiment.
[0059]
In the example of FIG. 10, the duplicate key data structure 102 includes a page 1002 that actually stores table data information and a page 1001 that stores a management entry 1003 having a pointer pointing to the page 1002.
[0060]
The management entry 1003 stored in the page 1001 has a pointer 1005 pointing to the page 1002 and a maximum value 1004 of table data information stored in the page 1002. Management entries 1003 in the page 1001 are sorted in the order of the maximum value 1004 of the table data information.
[0061]
As shown in FIG. 10, each page 1001 has a pointer to a page 1001 in which a management entry having maximum information larger than the maximum information 1004 in the management entry 1003 stored in the page 1001 is stored. The leaf entry 20 points to the page 1001 storing the management entry pointing to the page 1002 storing the smallest table data information.
[0062]
With the above data structure, it is possible to efficiently perform a search in the duplicate key data structure described with reference to FIGS.
[0063]
Further, new allocation / recovery of the page 1001 or the page 1002 can be easily realized by GET / PUT to the database buffer. The second log record at the time of new allocation / collection can easily perform the change continuation process in the cancellation process by adding the page identifier of the page 1001 or the page 1002 in addition to the configuration information shown in the example of FIG. is there.
[0064]
The processing of the flowchart shown above is executed as a program in the computer system shown as an example in FIG. However, the program is not limited to the program stored in the external storage device physically connected directly to the computer system as in the example of FIG. It can be stored in a storage medium readable and writable by a computer such as a hard disk device or a floppy disk device. Further, it can be stored in an external storage device connected to a computer system different from the computer system of FIG. 3 via a network.
[0065]
In this way, table data information with the same key can be accessed simultaneously by multiple processes without acquiring exclusion at the time of access, and can be canceled without restricting the access of normal processes Therefore, it is possible to provide high concurrency with cancellation processing for a plurality of processes that simultaneously access the same key.
[0066]
Since the duplicate key data structure is pointed from the leaf entry and stored and managed outside the leaf node, storage of leaf entries having more other keys within the leaf node in which the leaf entry having the duplicate key is stored Is possible. As a result, the storage capacity of the B-tree index is suppressed, and stable access performance can be provided.
[0067]
As described above, it is possible to provide a data management method using a B-tree index that has excellent concurrency and high storage efficiency even when there are a large number of duplicate keys.
[0068]
The data management method using the B-tree index is very effective for applications that use the B-tree index for data with a narrow range, such as “business transition state” in a workflow. Because the B-tree index used by the application has a narrow range, the data duplication related to one key is enormous, and the B-tree index is frequently updated due to changes such as the “business transition state”. is there.
[0069]
Although the B-tree has been described above, the present invention can be similarly applied when an index access to the same key occurs in an environment where a plurality of accesses are performed, and suitably performs an index access to the same key. It is possible to provide a data management method and apparatus capable of managing the data.
[0070]
【The invention's effect】
ADVANTAGE OF THE INVENTION According to this invention, the data management method and apparatus which can perform suitably the index access with respect to the same key in the environment where several accesses are performed can be provided.
[Brief description of the drawings]
FIG. 1 is a conceptual diagram of the present invention.
FIG. 2 is a diagram showing functional blocks of a database processing system having a B-tree index according to the present embodiment.
FIG. 3 is a diagram illustrating an example of a hardware configuration of a computer system according to the present embodiment.
FIG. 4 is a flowchart illustrating a processing procedure of index update processing of a B-tree index management unit 25 according to the present embodiment.
FIG. 5 is a flowchart illustrating a processing procedure of index update processing of a B-tree index management unit 25 according to the present embodiment.
FIG. 6 is a flowchart illustrating a processing procedure of index update processing of a B-tree index management unit 25 according to the present embodiment.
FIG. 7 is a flowchart illustrating a processing procedure of index update processing of the B-tree index management unit 25 according to the present embodiment.
FIG. 8 is a diagram illustrating an example of a B-tree index log record according to the present embodiment;
FIG. 9 is a flowchart illustrating a processing procedure of an index update result cancellation process (recovery processing procedure) of the B-tree index management unit according to the present embodiment;
FIG. 10 is a diagram illustrating an example of a configuration of a duplicate key data structure according to the present embodiment.
FIG. 11 is a diagram showing a configuration of a conventional B-tree index having many duplicate keys.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... B-tree index, 2 ... Database management system, 3 ... Database, 4 ... System log, 5 ... Table data, 6 ... Application program, 21 ... Logical processing part, 22 ... Physical processing part, 23 ... System control, 25 ... B-tree index management unit, 26 ... B-tree structure access processing, 27 ... duplicate key data structure access processing, 101 ... B-tree structure, 102 ... duplicate key data structure, 41 ... first log record, 42 ... second log record.

Claims (2)

In a data management apparatus in which one key manages identification information relating to data including the key as a data pointer, and uses a B-tree index to find a data pointer related to the key based on a key value .
A data structure for storing a plurality of data pointers related to a plurality of data including the same key as the one key, and a data structure different from a leaf node in which an index entry including the key is stored. Storage means having a duplicate data structure that is pointed from the index entry that includes the key and does not acquire exclusion when accessing a plurality of data pointers regarding a plurality of data that includes the same key;
When adding or deleting a data pointer related to a key, table data related information as information for adding or deleting the data pointer to the duplicate data structure storing a plurality of data pointers related to the key Means for obtaining a first log record including a pointer to the duplicate data structure and the key;
Means for updating the index entry containing the key stored in the leaf node;
Means for obtaining a second log record including information indicating that the data pointer has been added to or deleted from the duplicate data structure;
Means for adding or deleting the data pointer to the duplicate data structure;
When canceling the processing result that involves adding or deleting the above data pointer,
When the acquired log record is received and the received log record is the second log record, the second log record is retained, and the received log record is the first log record. If the second log record is held, it is confirmed that the addition or deletion of the data pointer to the duplicate data structure is completed. Means to judge,
Adding or deleting the pointer to the duplicate data structure by confirming that the second log record is not held even though the log record received in the above determination is the first log record Is determined to be not completed, the duplicate data structure indicated by the pointer to the duplicate data structure included in the first log record is accessed, and the table data information included in the first log record is added or A data management apparatus comprising means for deleting. A computer that records identification data related to data including the key as a data pointer and records a data management program using a B-tree index to find the data pointer related to the key based on the key value In possible recording media,
  A data structure for storing a plurality of data pointers related to a plurality of data including the same key as the one key,
The data structure is different from the leaf node in which the index entry including the key is stored, and the data structure is pointed to from the index entry including the key, and access to a plurality of data pointers regarding the plurality of data including the same key Have a duplicate data structure that does not get an exclusive at the time of
  When adding or deleting a data pointer associated with a key,
  For the duplicate data structure storing a plurality of data pointers associated with the key,
A function of acquiring a first log record including table data related information, a pointer to the duplicate data structure, and the key as information for adding or deleting the data pointer;
  The ability to update the index entry containing that key stored in the leaf node;
  A function of obtaining a second log record including information indicating that the data pointer has been added to or deleted from the duplicate data structure;
  A function of adding or deleting the data pointer to the duplicate data structure;
  When canceling the processing result that involves adding or deleting the above data pointer,
When the acquired log record is received and the received log record is the second log record, the second log record is retained, and the received log record is the first log record. If the second log record is held, it is confirmed that the addition or deletion of the data pointer to the duplicate data structure is completed. The ability to judge,
  Adding or deleting the pointer to the duplicate data structure by confirming that the second log record is not held even though the log record received in the determination is the first log record Is determined to be not completed, the duplicate data structure indicated by the pointer to the duplicate data structure included in the first log record is accessed, and the table data information included in the first log record is added or A computer-readable storage medium storing a data management program for causing a computer to execute a function to be deleted.

Images